Haddock and hoogle generation leaves behind some artifacts
stackage-server fetches hoogle databases and doesn't clean them up when it switches to new ones
Historically this didn't matter because (1) it was easy enough to manually clean up after curation and (2) stackage-server ran on ephemeral Kubernetes pods that would just get recycled if they ran out of disk space.
Now that all these processes run on the same server, and we need backups of the server's database, we should start managing the growth of disk usage.
Disk usage grows over time for a few reasons:
Historically this didn't matter because (1) it was easy enough to manually clean up after curation and (2) stackage-server ran on ephemeral Kubernetes pods that would just get recycled if they ran out of disk space.
Now that all these processes run on the same server, and we need backups of the server's database, we should start managing the growth of disk usage.